Microsoft (MSFT) is reportedly planning to significantly increase production of its next-gen AI chips in 2027 to persuade big cloud customers to use them, the Information reports
- Microsoft plans to unveil its new Maia 300 chip in autumn.
- Co. has also be in talks with TSMC (2330 TT) to secure manufacturing capacity for over 300k of its chips for delivery in 2027.
Hyperscaler custom-silicon programmes have followed a consistent pattern: initial deployment is captive, serving the firm's own workloads, and the harder test is convincing third-party cloud tenants to accept in-house accelerators over the incumbent merchant supplier's stack, where software ecosystem lock-in has historically been the binding constraint rather than raw silicon performance. Capacity reservations with the leading foundry of this scale signal intent well ahead of any revenue line, and in past episodes such wafer commitments have functioned as a leading indicator of how seriously a hyperscaler intends to contest the accelerator market rather than merely hedge its supplier. The merchant chip complex has tended to trade hyperscaler ASIC headlines as share-shift risk, though realised displacement has repeatedly lagged the headline given qualification cycles and software porting costs. What matters next is the unveil itself, performance and pricing disclosures against the merchant incumbent, and whether named cloud customers commit workloads, which is the tell that separates a strategic hedge from a genuine platform push. The foundry angle carries its own read-through on advanced-node capacity allocation among the big AI buyers.